Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Intel Corporation (INTC) -- Ignore unavailable to you. Want to Upgrade?

To: Dan3 who wrote (126882)	2/8/2001 12:34:09 AM
From: Scumbria	Respond to of 186894

Dan, P4's performance problems are not due to a lack of DRAM bandwidth, but rather imbalanced resources in the core. BTW: If multitasking is forcing a lot of DRAM traffic, that is a good indication that the caches are thrashing and therefore too many apps are running. Under those circumstances, it would be more efficient to run queueing software and serialize the applications. BTW: The multitasking argument is of course the last remaining claim of RDRAM, and as usual it doesn't hold any water. Scumbria

To: Dan3 who wrote (126882)	2/8/2001 1:51:41 PM
From: Tenchusatsu	Read Replies (1) \| Respond to of 186894

Dan, first of all, a server isn't just a glorified network router. So the memory bandwidth is used for more than just PCI-to-memory accesses. But it seems that you and I agree on this point. <And if P4 isn't positively drowning its memory bus with prefetch traffic, then why is dual channel PC800 necessary to equal the performance of a PIII or Athlon with single channel PC133? Why can't even dual channel PC600 supply enough bandwidth?> Someone pointed this out to me a year and a half ago. PC600 RDRAM has even worse latency than PC800. Also, as Scumbria points out, bandwidth isn't the problem in a Pentium 4 system. Prefetch serves two purposes. One, it's necessary because a P4 cacheline is 128 bytes long, but the maximum burst length of a single FSB transaction is 64 bytes. So when the first transaction is seen on the FSB, the second transaction is anticipated to be right behind it. And two, prefetch helps make use of the extra memory bandwidth that would normally be wasted in the middle of non-bandwidth intensive tasks. It is unclear how useful prefetching really is on most applications. But hey, if the bandwidth is available, why not leave it on and hope that it can buy you an extra bit of performance? (By the way, the prefetch logic was borrowed from the 840 Carmel chipset, which is no surprise since the 850 is pretty much an upgraded 840.) <I'd be interested in an informed criticism but if all you can come up with is "Wrong" then never mind.> You pretty much assumed that false statements were true without even thinking about whether they made sense or not. All they deserved was "Wrong." Tenchusatsu